A model with a context window, or a model with your customer graph

A chat reasons over the material available in its prompt or selected by its retrieval setup. As the corpus grows, the user has to decide what to upload, summarize, or retrieve for each analysis.

Poth runs against connected, permission-bounded customer sources and returns cited evidence with the answer. Readers can inspect what supports the result and identify where more evidence may be needed.

In a one-off chat workflow, a citation is still generated output unless the surrounding application verifies it against the retrieved source.

Poth validates citation markers against the evidence supplied to an answer and displays cited-source cards with excerpts. That makes the supporting material inspectable without presenting an unverified benchmark as proof.

Joining what customers say to what they do is bookkeeping, not reasoning. In an ad hoc chat workflow, those speaker, account, subscription, and usage mappings have to be supplied or rebuilt for each analysis.

Poth maintains that mapping persistently, so an analysis can join account behavior to cited conversations without rebuilding those identities for each run.

A one-off chat analysis is easy to start, but the context, joins, and definitions often live in that conversation instead of a shared customer-data layer.

Poth keeps the connected source inventory and customer graph available across analyses. Ask Poth conversations are saved per user; generated answers do not become shared claims automatically.

Ad hoc runs can bucket feedback differently unless the workflow supplies and preserves a taxonomy. That is manageable for a one-off and risky for "is onboarding friction down since the fix?"

The company brain persists objects and relationships built from connected evidence, so a repeat analysis can start from maintained company context instead of an empty conversation.